Claude Code Audit

Most Claude Code projects accumulate quiet problems: a CLAUDE.md that's too long to be read, skills that never trigger because their description is too vague, hooks that silently fail, or a .env file one git add away from being committed. claude-code-audit catches all of this in one pass.

It reads your CLAUDE.md, skills, hooks, MCP server config, .gitignore, and settings.json — then produces a severity-ranked report (CRITICAL / WARNING / INFO) with plain-language explanations, concrete fixes, and ready-to-paste Claude Code prompts so you can resolve each finding immediately. What it audits:

CLAUDE.md — length, token cost, missing sections, domain knowledge that belongs in a skill Skills — frontmatter completeness, description quality, double-nesting bugs Hooks — destructive commands, broken paths, overly broad matchers MCP servers — context overhead, unsynced configs Sensitive files — .env, credentials, and API keys not covered by .gitignore Project hygiene — settings.json permissions, directory structure

Output: header, score, findings with priority order, and a single closing sentence naming the most important fix.